Tropospheric dry layers in the tropical western Pacific : comparisons of GPS radio occultation with multiple data sets

We use GPS radio occultation (RO) data to investigate the structure and temporal behavior of extremely dry, high-ozone tropospheric air in the tropical western Pacific during the 6-week period of the CONTRAST (CONvective TRansport of Active Species in the Tropics) experiment (January and February 2014). Our analyses are aimed at testing whether the RO method is capable of detecting these extremely dry layers and evaluating comparisons with in situ measurements, satellite observations, and model analyses. We use multiple data sources as comparisons, including CONTRAST research aircraft profiles, radiosonde profiles, AIRS (Atmospheric Infrared Sounder) satellite retrievals, and profiles extracted from the ERA (ERA-Interim reanalysis) and the GFS (US National Weather Service Global Forecast System) analyses, as well as MTSAT-2 satellite images. The independent and complementary radiosonde, aircraft, and RO data provide high vertical resolution observations of the dry layers. However, they all have limitations. The coverage of the radiosonde data is limited by having only a single station in this oceanic region; the aircraft data are limited in their temporal and spatial coverage; and the RO data are limited in their number and horizontal resolution over this period. However, nearby observations from the three types of data are highly consistent with each other and with the lower-vertical-resolution AIRS profiles. They are also consistent with the ERA and GFS data. We show that the RO data, used here for the first time to study this phenomenon, contribute significant information on the water vapor content and are capable of detecting layers in the tropics and subtropics with extremely low humidity (less than 10 %), independent of the retrieval used to extract moisture information. Our results also verify the quality of the ERA and GFS data sets, giving confidence to the reanalyses and their use in diagnosing the full four-dimensional structure of the dry layers.


Introduction
Water vapor is the most important greenhouse gas in the troposphere, yet it is still the parameter with the highest uncertainty in weather and climate models.All current humidity observation techniques have limitations for analyzing global water vapor fields.For example, nadir-viewing satellite instruments (e.g., infrared (IR) or microwave (MW) sensors in space) are restricted by their low vertical resolution.IR sounders cannot observe within and under clouds.Radiosonde (RS) coverage is sparse or nonexistent over the open oceans.The radio occultation (RO) method does not suffer from these limitations, but water vapor information can only be derived by using a combination of RO data and information on temperature from another source (observations or model).In addition, RO observations have significant errors in the lower tropical troposphere under superrefraction conditions (Sokolovskiy, 2003).
A number of studies have shown good agreement between RO and RS moisture profiles (e.g., Kishore et al., 2011;Ho et al., 2010).These climatological studies focus on comparisons of RO with the global radiosonde network rather than extremely dry air (relative humidities less than 10 %).In this paper we study the ability of RO to measure extremely dry Published by Copernicus Publications on behalf of the European Geosciences Union.
air in the tropics and subtropics, using for comparison highresolution aircraft profiles, radiosondes, IR satellite data, and the ERA (Berrisford et al., 2011).
Dry regions of the tropical and subtropical lower and mid-troposphere have a strong radiative impact on the climate system through their ability to radiate heat to space, preventing a "runaway greenhouse effect" (Pierrehumbert, 1995).Furthermore, they suppress deep convection (Brown and Zhang, 1997), are connected to cumulus congestus cloud top heights (Johnson et al., 1996), and affect boundary layer height and humidity (Parsons et al., 2000).
A number of studies have addressed the so-called dry intrusions in the normally moist mid-and lower troposphere of the tropical western Pacific.They were first investigated during TOGA-COARE, the Coupled Ocean-Atmosphere Response Experiment of the Tropical Ocean and Global Atmosphere project (Webster and Lukas, 1992).Mapes and Zuidema (1996), using soundings from TOGA-COARE, found that dry layers are generally too dry and not warm enough to be interpreted as adiabatic displacements within the tropics.Instead they suggest a subtropical origin.Dry layers typically have strong horizontal and vertical moisture gradients and sharp temperature inversions at the lower edge.They are stabilized by radiative cooling of the underlying moist air and heating of the dry air layer, thus inhibiting convection.Cau et al. (2005) investigated the radiative impact and origin of dry intrusions observed by RS profiles in the tropical western Pacific using 40-year European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis (ERA-40) wind and humidity data.They showed an outgoing longwave radiation increase of 3 W m −2 per 100 hPa for dry intrusions with relative humidities of less than 20 %, almost independent of altitude.They pointed out the importance of the humidity distribution in a climate change scenario, considering that outgoing longwave radiation is more sensitive to small humidity perturbations in dry environments than in moist regions.Cau et al. (2005) also pointed out that cloud occurrence above or below the dry intrusion reduces the radiative impact.Furthermore, they used back trajectories to show that most dry events were associated with air descending from the extratropics.In their follow-up study, Cau et al. (2007) evaluated the origins of dry air in the tropics and subtropics using trajectory simulations for ERA-40 data for January 1993.They found four major transport mechanisms: (1) the descending branch of extratropical baroclinic waves; (2) the equatorial flank of the jet around subtropical anticyclones; (3) transport at regions of minimum subtropical jet strength via equatorward descent across the jet exit; and (4) dry air centering in the upper troposphere between regions of deep convection (see also Fig. 9 in Cau et al., 2007).
Regarding dry layer occurrence, Casey et al. (2009) created a 5-year climatology on dry layers between 600 and 400 hPa over deep convective regions of the tropical oceans using AIRS (Atmospheric Infrared Sounder) data.Their re-sults show large spatial and seasonal variability for different ocean basins, pointing out the limits of applying case study trends to the whole basin.
Finally, Randel et al. (2016) performed a detailed comparison between aircraft measurements from the CONvective TRansport of Active Species in the Tropics (CONTRAST) experiment and GFS meteorological analyses, demonstrating that the analyses accurately capture the behavior of subtropical dry layers.A global climatology from GFS data shows that the dry layers are a ubiquitous feature of the subtropics, with maximum occurrence frequency in the winter hemisphere (linked to the strongest subtropical jets).The subtropical dry layers are highly correlated with enhanced ozone in both hemispheres, arguing for a source in the extratropical upper troposphere-lower stratosphere (UTLS).
A number of studies confirmed the capability of RO measurements to monitor the dry atmosphere (above around 8 km; Foelsche et al., 2008Foelsche et al., , 2009) ) and for climate change detection (Leroy et al., 2006;Ho et al., 2009;Steiner et al., 2011;Ho et al., 2012).RO data feature inherent high accuracy and precision, high vertical resolution (100 to 200 m), all-weather capability, and long-term stability (Anthes, 2011), making them highly valuable for studying a large number of atmospheric phenomena.Vergados et al. (2015) studied the spatial variability of relative humidity (RH) from RO compared to ECMWF and MERRA (Modern-Era Retrospective analysis for Research and Applications), focusing on time-averaged seasonal behavior; these comparisons suggest an overall reliable behavior of RO-derived humidity fields.So far no study has focused on RO and extreme dryness (RH < 10 %).
RH is computed from measured water vapor pressure and saturation water vapor pressure over liquid or ice, depending on the temperature.The liquid formulation is used according to Murphy and Koop (2005).
This paper is structured as follows: in Sect. 2 we summarize the RO technique, the CONTRAST field campaign, and all other data sets we used.In Sect. 3 we show some example profile comparisons, explain features of dry layers, and discuss the contributions of a priori (first-guess) data and RO observations in the one-dimensional variational (1D-Var) retrieval.Section 4 focuses on one specific case in detail.In Sect.5, we give a short overview of the results using all collocation pairs available.Section 6 discusses the global occurrence of dry layers derived from RO data.Section 7 provides a summary and conclusions.

The RO method
The RO method (Melbourne et al., 1994;Hajj et al., 2002;Kuo et al., 2004) is a limb-sounding technique that provides near-vertical profiles of atmospheric refractivity N. The re-lation of N to atmospheric temperature T , pressure p, and water vapor pressure e can be approximated by the Smith and Weintraub (1953) formula: Additional terms accounting for contributions from liquid water and the ionosphere can be neglected or are accounted for earlier in the retrieval.In the so-called dry air retrieval, the "dry temperature" is computed using Eq.(1) under the assumption e = 0.For a detailed retrieval description, see Kursinski et al. (1997).
Water vapor pressure e in Eq. ( 1) cannot be determined from an observed N without ancillary temperature data from some other source (either observations, a model, or analysis).The two common techniques for this calculation are discussed in Appendix A, as well the influence of the ancillary data in the 1D-Var retrieval (Appendix B).
For this study, we downloaded data from CDAAC 1 (COS-MIC Data Analysis and Archive Center) for the RO missions COSMIC (Constellation Observing System for Meteorology, Ionosphere and Climate; reprocessed data cos-mic2013), GRACE (Gravity Recovery and Climate Experiment; post-processed), Metop-A (Meteorological Operational Polar Satellite A; reprocessed data metopa2016), Metop-B (Meteorological Operational Polar Satellite B; post-processed), and TerraSAR-X (post-processed).
CDAAC provides profiles of physical parameters, which are derived by using a 1D-Var retrieval (COSMIC, 2005)).In the 1D-Var, ERA profiles (interpolated to the location and time of the RO measurement) are used as the initial (first-guess or a priori) temperature and moisture profiles in the iteration procedure.Furthermore, CDAAC also provides these a priori profiles, and collocated profiles from other (re-)analyses.In this study we use these RO-collocated profiles from ERA and GFS for comparisons.ERA vertical profiles are provided at 25 hPa steps between 1000 and 750 hPa and at 50 hPa steps between 750 and 300 hPa, yielding a total of 19 levels below 300 hPa.GFS analyses are given at 50 hPa steps from 1000 to 300 hPa and at 975 and 925 hPa additionally, yielding a total of 16 levels below 300 hPa.

The CONTRAST experiment
The CONTRAST experiement was conducted over the western Pacific warm pool region during the season characterized by intense convective storms to study the impact of deep convection on chemical composition and ozone photochemical budget (Pan et al., 2017).The experiment was conducted from Guam (13.5 • N, 144.8 • E) using the NSF/NCAR Gulfstream V (GV) research aircraft during January and February 2014.During the campaign, 16 research flights were conducted.Most research flights included several vertical pro-1 http://cdaac-www.cosmic.ucar.edu/cdaac/files (covering altitudes from 0.1 to 15.2 km), and together with take-offs and landings at Guam there were over 80 vertical profiles obtained during the experiment.We use the aircraft observations of temperature, pressure, and water vapor pressure to derive high-resolution vertical profiles of RH.
Water vapor was measured by the Vertical Cavity Surface Emitting Laser (VCSEL) hygrometer (absolute concentration of water vapor in molecules per cubic centimeter).It is designed to work throughout the troposphere (and also the lower stratosphere) and has an accuracy of ±6 % mixing ratio +0.3 ppmv and a precision of ≤ 3 % (see Zondlo et al., 2010, for details).Temperature was measured by two Harco heated total air temperature sensors (estimated accuracy: 0.5 • C; precision: < 0.01 • C), pressure was measured using the Paroscientific, Ltd. sensor model 1000 transducer (accuracy: 0.1 hPa, precision: < 0.01 hPa)2 .From the CON-TRAST netcdf files, the variables used for T , e, and p are ATX, EW_VXL, and PSXC.

ERA-Interim reanalysis
In addition to the RO-collocated ERA profiles (as described in Sect.2.1), we downloaded ERA-Interim reanalysis fields from European Centre for Medium-Range Weather Forecasts ( 2009) for the time range of the CONTRAST experiment.They are available every 6 h at 00:00, 06:00, 12:00, and 18:00 UT.We use the data on the lowest 27 levels, from 1000 to 100 hPa.ERA uses a 4D-Var method and assimilates RS observations, AIRS radiances, and RO bending angles, among other in situ and satellite data (Dee et al., 2011).

Radiosonde, AIRS, and MTSAT-2 observations
RS data from Guam were downloaded from NOAA3 .Data are available at approximately midnight and noon UT (10:00 and 22:00 local times, respectively).Measurements are taken at standard pressure levels and significant thermodynamic levels.To convert the pressure grid to altitude, we used a constant temperature gradient of 6.5 K km −1 .
AIRS is a cross-track scanning instrument on the NASA Aqua satellite.Its sun-synchronous, near-polar orbit is designed to cross the Equator from south to north at 13:30 local time.The NASA Goddard Sciences Data and Information Center provides AIRS retrieved data products, such as profiles of physical parameters (temperature, humidity) and trace gas constituents, on a daily basis.We downloaded the version 6 level 2 standard retrieval data4 .
MTSAT-2 is a geostationary satellite located over Australia, East Asia, and the western Pacific, operated by the T. Rieckh et al.: Tropospheric dry layers in the tropical western Pacific Japan Meteorological Agency.Detailed information can be found at Knapp (2008).MTSAT-2 carries an imaging telescope, backed by detectors for five wavelength channels.
We use data from the infrared channel (10.3 to 11.3 µm) and the water vapor channel (6.5 to 7 µm).We downloaded MTSAT-2 data from NOAA5 .
The sampling characteristics of the different observation and model data sets compared in this study vary greatly.The radiosonde and aircraft data are essentially point measurements.Radiosonde measurements are taken on standard pressure levels and significant thermodynamic levels, which results in the vertical resolution varying strongly within the profile (from less than 20 to almost 1000 m).The vertical resolution of the aircraft measurements is around 10 m.The horizontal footprint of the radio occultation profiles is ∼200 km while the vertical resolution is 100 to 300 m (Anthes, 2011).The AIRS level 2 products are reported on 28 standard pressure levels between 1100 and 0.1 hPa.They have a horizontal resolution of 50 km6 and a vertical resolution of ∼1 km for temperature and ∼2 km for humidity7 .For the two models used in this study, we used GFS and ERA interpolated to the time and location of the RO profile.Furthermore, we used the ERA field, for which the horizontal footprint is given by the horizontal grid size (0.7 • × 0.7 • , about 78 km × (68 to 78) km, depending on the latitude).The vertical resolution is given by the model pressure levels every 25 or 50 hPa (resulting in a vertical resolution between 200 to 1000 m).
Because both radiosondes and the aircraft measurements are essentially rapid-response point values and have high vertical resolution, they are capable of measuring turbulence and small-scale horizontal features (such as individual clouds).AIRS, RO, and models, in contrast, represent averages over much larger horizontal and vertical scales of observation (larger horizontal and vertical footprints).Thus different volumes of air are sampled and compared, leading to representativeness errors or differences due to their different horizontal and vertical footprints, especially when measuring fields with high temporal and spatial variability such as water vapor and relative humidity.
In addition, all the observations and model data occur at different locations and times.The RS, RO, and aircraft observations occur at different horizontal positions in the vertical (can be ∼100 to 200 km) as the balloons, satellites, and aircraft move during the "vertical" sounding.All of these differences make comparisons challenging, adding to the uncertainties associated with each individual data set, and must be considered when interpreting the results.

Collocating CONTRAST and RO profiles
From the 16 research flights, we extracted 75 profiles that extend over at least 6 km altitude and are within the region 5 • S to 20 • N latitude, 130 to 170 • E longitude (Fig. 1).We tried different criteria for maximum time and distance for collocating RO with CONTRAST profiles.We made these comparisons only for the lowest 10 km, since RO observations do not provide reliable moisture values at altitudes above about 8 km (Kishore et al., 2011).Aircraft-RO profile pairs with less than 4 km overlap in the vertical were discarded (CONTRAST profiles are mostly limited by their maximum altitude; RO profiles are limited by their minimum altitude).The time and space coincidence criteria tested included 3 h and 600 km, 12 h and 300 km, and 24 h and 200 km, yielding 37, 41, and 24 profile pairs, respectively.Note that the shortest time windows in these criteria correspond to the longest spatial intervals; this is done to ensure enough pairs in sets matching each of the criteria.Using a criterion of the shortest time and space separation would not yield enough pairs to make the results as meaningful.As shown in Sect. 5 below, all three of our criteria gave similar results; thus we only show results for the 3 h and 600 km criteria.
3 Individual profile comparisons Figure 2 shows an example of a dry layer sampled by the CONTRAST research flight 2 profile d (rf02d) and the collocated RO profile (Metop-A) for the parameters RH, T , q, and N .RH (upper left) shows a typical dry layer structure.There is a strong drop in RH at the bottom of the layer (at ∼2.5 km) from > 80 to < 10 %.All profiles (CONTRAST, RO, ERA, and GFS) show extremely dry air above 4 km.This layer is particularly thick and the RH remains below 20 % up to 10 km.The RO profile shows a remarkably similar shape when compared to CONTRAST, including the sharp humid- ity gradient at the bottom of the dry layer.Both ERA and GFS show a less sharp vertical moisture gradient, partly due to lower vertical resolution (RO provides 80 levels between the surface and 8 km, while ERA and GFS provide 18 and 15 levels, respectively).Profile differences are largest in the lowest 2.5 km, where RO and ERA RH are up to 20 % lower than the CONTRAST and GFS RH.
The aircraft T profile (Fig. 2, upper right) shows a strong inversion at the altitude of the bottom of the dry layer, as has been described by, e.g., Mapes and Zuidema (1996).Neither ERA and GFS nor RO not detect this strong inversion.RO generally has the capability to resolve such strong inversions in the middle and upper troposphere (Anthes, 2011).We conclude that RO not showing the T inversion could be explained by two factors: (1) CONTRAST and RO have different N values at this altitude (see Fig. 2, lower left).This implies that there has to be a difference in T and/or e at this altitude, too.(2) The 1D-Var retrieval generally produces an RO T close to the first-guess T , which does not show the inversion, and changes mainly e in the adjustment of the firstguess N towards the measured RO N.
Specific humidity q (lower right) shows extremely dry conditions above 2.5 km, but in less extreme cases dry layers are harder to detect using this parameter since it generally de-creases exponentially with altitude.Thus we mainly use RH to investigate dry layers.
Figure 3 shows two more examples for CONTRAST-RO pairs (Metop-B and COSMIC-FM6).In the left panel, CON-TRAST depicts a dry layer from 2.5 to 5.5 km, a relatively moist layer from 5.5 to 6.5 km, and very dry air from 6.5 to 8 km.The RO RH shows a very similar overall structure.Major differences are again in the lowest 2 km, where CON-TRAST, GFS, and ERA RH are up to 20 % higher than RO RH, and between 2.5 and 4 km, where CONTRAST RH is up to 20 % lower.Furthermore, both GFS and ERA miss the 1 km thick moist layer around 6 km.
Figure 3, right, shows profiles with two drier layers in the mid-troposphere (at 3 and 5 km), but no extreme dryness.Both ERA and GFS show the correct overall shape of the RH profile, but they are often 20 to 30 % drier or moister than the CONTRAST RH.Again, RO captures more vertical structure than the models.
Generally both of these examples show how well the RO RH profile structure agrees with the one from CONTRAST.We found that many CONTRAST vs. RO profile pairs matched very closely, but some of the pairs showed RH differences of more than 60 % at certain levels.In this section we look into one of these cases in more detail to help explain these strong discrepancies.We consider two specific profiles, one measured by CONTRAST starting at Guam on 19 February 2014, 17:00 UT (Fig. 4, top), and one landing at Guam a few hours later (20 February 2014, 00:22 UT; Fig. 4, bottom).
For both cases, the left side shows the profiles for CON-TRAST, the closest collocated RO, the RO-collocated ERA and GFS profiles, a RS launched at Guam (RS G ), and two AIRS profiles (one closest to Guam, labeled as AIRS G ; and one closest to the RO profile, labeled as AIRS RO ).Table 1 lists the time differences and distances between these profiles.
The right plots show the ERA RH field at 500 hPa, closest in time to the respective CONTRAST profiles.Also shown are the ERA ozone values and winds.The location of Guam is marked by a white X.CONTRAST, RO, RS, and AIRS profiles are labeled.Additional squares indicate other RO measurements.The color filling of the RO symbols (white squares) varies with the RH of the RO observation at this level, with the same color code as the ERA RH analysis (color bar).It is noteworthy that in almost all the cases the colors (and hence RH) of the RO observations agree very closely with the ERA values.
In Fig. 4, top left, all profiles show a deep dry layer.The depth of the layer varies slightly between the data sets.For CONTRAST, RS, and RO the depth with RH ≤ 10 % varies between 5 and 6.5 km.Both the ERA and GFS profiles and both AIRS profiles show a less sharp transition from moist to dry.The models also show a generally thinner dry layer.Major differences between the data sets occur below 2 km, where RO is significantly drier than all other data sets.The GFS profile agrees with the drier RO profile down to about 1 km and then strongly increases in RH.Sometimes superrefraction can cause a dry bias in RO profiles in the lowest few kilometers, which is probably the case in this particular RO profile.
Figure 4, top right, shows the ERA RH field at 500 hPa for the whole region.The large-scale region of very dry air extends from 110 to 180 • E with a width of 1600 to 2200 km.It also shows the high horizontal variability of moisture.In some areas, extremely strong horizontal RH gradients occur, which clearly mark the edge of the dry air mass.In these areas RH can increase from less than 20 % to more than 70 % within 100 km.And since this is a model field, in which the gradients are likely to be smoother than the real atmosphere, the actual gradient could be even sharper.In this figure, it is also clearly visible that all the profiles from the left panel are located in the same air mass.Although ozone is not a focus of this study, we include contour lines of ozone in Fig. 4 (right panels) to illustrate that the ERA ozone fields show, in general, a coincidence of high ozone values with very dry air, suggesting the origin of at least some of the dry air from the lower stratosphere.
Next we consider the same region about 6 h later (Fig. 4, bottom panels).The profiles (left) show that the dry layer persists at Guam and is even deeper for CONTRAST.The RS profile and the AIRS profile at Guam still show a very deep dry layer.However, the RO shows only a very shallow dry layer, and RH increases from less than 20 % to more than 80 % between 3.5 and 4.5 km.The RO-collocated ERA and GFS profiles lack the dry layer entirely, having their RH minima at 45 and 50 %, respectively.They also both increase above 3.5 km to around 80 %.The AIRS profile at the RO location starts much drier than RO, ERA, or GFS at the surface.Overall, the AIRS profile has the same shape as RO, ERA, and GFS.It has weaker dry-moist transitions, similar to the (re-)analyses, and it stays drier than the other data sets Table 1.Distances and time differences between different the collocated profiles of rf13a and rf13d at the lowest point for each profile.Note that both the aircraft and RS profile will move away from Guam with higher altitudes and that RO profiles are also not completely vertical, especially in the lower and mid-troposphere.The lower right panel in Fig. 4 shows that the RO (and thus also the RO-collocated ERA and GFS) profiles are located in the moist air mass, just a few tens of kilometers from the edge of the moist-dry boundary, with much higher moisture  values above 3 km.This explains the very different profiles in the left panel.Finally, Fig. 5 shows satellite images from MTSAT-2 for approximately the same time as the ERA field in Fig. 4, bottom right.The satellite images are cropped to the latitude and longitude range as ERA (as much as the different projections allowed).The Guam and RO locations are marked by an X.
The left panel depicts the brightness temperatures from IR in an atmospheric window (at 10.3 to 11.3 µm).It is derived from terrestrial IR radiation emitted by the Earth, cloud tops, and the atmosphere.Color enhancement shows the high, cold cloud tops south of Guam.Conditions are clear around Guam and the CONTRAST profile, but the RO profile is located in cloudy air.
The right panel shows the image from the MTSAT-2 water vapor channel.It depicts brightness temperatures derived from the water vapor emission spectrum between 6.5 and 7 µm.Higher amounts of water vapor absorb more radiation, which is re-emitted.Thus regions with high amount of water vapor, especially in the upper troposphere or above clouds, will have a higher brightness temperature.When compared to ERA (Fig. 4, bottom right), we see that the low RH re-gion has a much lower brightness temperature than the moist region south of Guam.

Results of all collocations
To get a general overview of profile pair differences we computed statistics (not shown) and created scatter plots for all collocated profiles.Since humidity has a strong spatial variation (as seen in Fig. 4, right) and also varies strongly with time, we tested different sets of collocation criteria (described in Sect.2.5).CONTRAST profiles are smoothed by taking 60 s averages in each profile (resulting in about 300-500 m vertical resolution).The RO profiles are interpolated to the related CONTRAST altitude grid.We did comparisons for RO and CONTRAST, ERA and CONTRAST, and GFS and CONTRAST.Generally, all three sets of collocation criteria and all data set comparisons show similar results for both the profile statistics and scatter plots.To illustrate the similarities, we show the mean, root mean square (RMS), and Pearson R correlation coefficient for RH for RO-CONTRAST, ERA-CONTRAST, and GFS-CONTRAST differences for all three criteria in Table 2.We chose RH due to our focus on tropospheric moisture in this paper.Furthermore, RH is the only parameter without some inherent vertical correlation due to a general decrease with altitude (i.e., N, T , and q).
Figure 6 shows the scatter plots for refractivity N, temperature T , specific humidity q, and RH for RO compared to CONTRAST.The color of the dot indicates the altitude from 0 to 10 km as shown in the color bar.The black solid line is the fitted linear regression, and the dashed black line indicates perfect agreement between the two data sets (slope = 1, intercept = 0).
The upper left panel depicts the N comparison between RO and CONTRAST.N decreases exponentially with altitude, so the spread is larger at lower altitudes (blue) than at higher altitudes (red).Overall, there is a high correlation between the two data sets, and the fitted regression (solid black line) agrees well with the line of perfect agreement (dashed black line).
T (top right) also shows very good agreement (high correlation and little spread).We found a small warm bias of 0.5 to 1.5 K in CONTRAST temperatures when compared to any of the other data sets (RO, ERA, GFS).To test how much influence the collocation criteria have, we interpolated the ERA field to the CONTRAST profile location (spatial difference for latitude and longitude < 0.5 • ; time difference less than 30 min), which yielded a slightly smaller, more uniform bias.We conclude that there is likely a small T bias in the aircraft temperatures, possibly because of the effects of solar radiation as most of the flights occurred during the daytime.
The specific humidity q is depicted in the bottom left panel.The spread appears larger for low values (very dry air, < 1 g kg −1 ), but the scales in this panel are logarithmic, which makes small differences of dry values appear larger.RO q values are biased positively compared to CONTRAST for low q values and biased negatively for high q values.Comparison on q between CONTRAST and GFS in Randel et al. (2016) shows a similar bias.Because of the highly accurate aircraft water vapor and temperature measurements and the very small scale of the observation (essentially a point observation), the CONTRAST measurements are capable of detecting extremes of dry and moist air more frequently than RO observations or model estimates, whose data represent averages over larger scales.
RH plots (bottom right) are highly scattered and have a lower correlation coefficient of around 0.78 with a bias and large spread in the data sets.The moist bias of RO for very dry air was already noted in the paragraph above.Thus CON-TRAST shows a much higher frequency of very low RH values than both RO (Fig. 6 lower right) and ERA (Fig. 7, left).The large spread can be explained by several factors: (1) RH is sensitive to both small variations in T and q, and thus representativeness differences or errors of both T and q contribute to differences in RH; (2) RH does not have a vertical profile with a mean structural or climatological variation in the vertical as N, T , q do (with an overall decrease with altitude); and (3) RH can undergo extremely strong changes in the vertical (80 % RH change over 1 km in Fig. 2), which leads to strong differences between two data sets if they do not capture this jump of RH at exactly the same altitude.
Furthermore, comparisons using 8 years of RS (two stations) and RO data using very tight collocation criteria (1 h, 100 km; figure not shown) also showed highly scattered data for RH, which indicates that collocation errors are not the dominant factor in the large scatter between RO and CON-TRAST relative humidities.
Figure 7, left, compares ERA and CONTRAST RH, which shows a similar strong scatter as RO and CONTRAST.This suggests that the high variability in the CONTRAST data set also plays a role in the strong scatter of RH.Both the RO and ERA data sets are horizontally smoothed: RO shows an average over about 200 km (limb-sounding), and ERA is interpolated from the nearest grid points to the location of the profile (horizontal resolution is < 0.7 • in latitude and longitude).Finally, Fig. 7, right, compares GFS to ERA (this comparison is done on the ERA pressure grid).The correlation between the two analyses is high, but the scatter is surprisingly large considering these are smooth model data sets.This shows how highly variable RH is.

Global distribution
Having shown that the RO observations are capable of detecting extremely dry layers in the tropical western Pacific region, we carried out a global climatology of dry layers using only RO data.AIRS data have been used before to find tropical dry regions within areas of convection (OLR < 240 W m −2 ; Casey et al., 2009); however, AIRS cannot provide reliable measurements below clouds.
Figure 8 shows the global occurrence (percentage of observations) for RH < 10 % on the 320 K potential temperature level for December-January-February (DJF) and June-July-August (JJA) 2014.We use the 320 K level because dry air travels from the stratosphere into the troposphere along isentropes.The 320 K level is at about 600 hPa or 4.5 km in the tropics and slopes to higher altitudes (∼9 km) in the extratropics.So the 320 K level represents the mid-troposphere in the tropics and the lower stratosphere in the extratropics.
Figure 8, left, shows the months DJF.Between 5 and 25 • N, almost the entire latitude band shows an occurrence of dry air for 50 % of the time or more.The only break in the band is off the west coast of North and Central America.In some regions, dry layer occurrence is as high as 75 %, e.g., in parts of the Atlantic or near India and the Arab peninsula.Guam is located just on the edge of the band of high frequency of dry layer occurrence.In the SH, two regions with very strong occurrence are easy to identify: one off the west coast of South America, and one in the southern Atlantic Ocean.
In JJA (Fig. 8, right), dry layers occur over smaller regions of the world, but with a much higher frequency.In the SH, the entire region from mid-Atlantic via Africa and the Indian Ocean to Australia shows a frequency of occurrence of 80 to 100 % of the time.In the NH, only a region in the northern east Pacific and the northeast Africa/eastern Mediterranean region show moderate to high occurrence.Overall, dry layers occur throughout the year in some regions of the SH, but there is also a seasonal cycle with higher occurrence in the winter hemisphere.This behavior is consistent with the dry layer climatology derived from ERA (results not shown) and from GFS data in Randel et al. (2016).

Conclusions
In this study we compared GPS RO profiles with multiple data sets to examine their ability to detect extremely dry layers in the lower and mid-troposphere in the tropics and subtropics.The comparisons were made in the tropical western Pacific, making use of the field experiment in that region during January and February 2014.We used data from the NSF/NCAR GV research aircraft as a reference for RO profiles.This is a challenging comparison, since the aircraft provides high-resolution profiles consisting of point measurements (capturing a lot of variability), whereas the RO technique is a limb-sounding technique (measuring the limbintegrated value at each profile level).Furthermore, we used RS and AIRS profiles, as well as data from the GFS model and ERA reanalysis.Our main findings and conclusions are as follows: 1. Radio occultation is capable of detecting layers with very low humidity in the lower and mid-troposphere, despite the fact that the water vapor pressure is only a small fraction of the measured refractivity in dry air.Comparing RO to other types of observations also shows that the structure and intensity of dry layers are captured quite accurately.
2. Both simple and 1D-Var RO water vapor retrievals yield similar results, demonstrating the ability of RO to retrieve water vapor profiles with fine-scale vertical structure similar to that of the aircraft profile (Appendix A).
3. There is significant information content in the RO water vapor retrievals, and the 1D-Var retrieval does not depend strongly on an accurate first-guess (a priori) information (such as from ERA).However, poor a priori information for water vapor may have an effect of several Kelvin on the retrieved temperature (Appendix B). 5.Both the GFS and ERA analyses show the overall correct structure when compared with CONTRAST aircraft observations and RO.They often exhibit less small-scale vertical variations or sharp vertical gradients, probably due to lower vertical resolution.
6.A detailed case study illustrated how strong horizontal moisture gradients (more than 50 % RH change within 100 km) can yield nearby profile pairs that strongly disagree, even though they are close in space and time.
7. Globally, dry layers occur throughout the year, mainly between 10 and 30 • N and S. Occurrence frequency is stronger in the winter hemisphere.The independent RO data confirm results from both ERA and GFS, which show a very similar seasonal occurrence of dry layers.
In summary, these diverse data sets show generally good agreement in spite of their large differences in sampling characteristics and technologies.
The CONTRAST data are available at https: //data.eol.ucar.edu/dataset/383.012.The ERA data are available at the NCAR Research Data Archive: http://rda.ucar.edu/datasets/ds627.0/.The radiosonde data are available at https://www.ncdc.noaa.gov/data-access/weather-balloon/integrated-global-radiosonde-archive.The AIRS data have been downloaded from ftp://airsl2.gesdisc.eosdis.nasa.gov/ftp/data/s4pa/Aqua_AIRS_Level2/AIRS2RET.006/.MTSAT-2 data have been downloaded from http: //www.ncdc.noaa.gov/gibbs/availability/2014-02-20.(p <200 hPa) from observed RO refractivity N will provide reliable estimates, because at these altitudes water vapor contributes very little to the refractivity; i.e., the so-called "wet" term in Eq. ( 1) is less than 1 % of the first, or "dry" term as noted by Wang et al. (2013).There is simply not enough information on water vapor pressure in refractivity at these altitudes to retrieve accurate estimates of water vapor.For example, the mean hurricane-season tropical atmosphere (Dunion and Marron, 2008) gives the following value of T at 200 hPa (about 12.4 km altitude): T = −54.6 • C (218.6 K).The sat-uration vapor pressure at this temperature is 0.04 hPa; thus for 100 % relative humidity, the dry term for N is 71.0 and the wet term is 0.31, or 0.4 % of the refractivity value.For the above values, the relationship between errors in e and T (Eq.A1) gives e = 0.042 T (hPa).So a temperature error (difference) of 0.5 K gives a difference in e of 0.021 hPa, which is more than 50 % of the saturation vapor pressure at this temperature.This example illustrates the difficulty in calculating water vapor pressure and relative humidity in the upper troposphere.Vergados et al. (2014) estimated the re- trieval errors in specific humidity at different pressure levels (925, 850, 700, 500, and 400 hPa).They estimated that in the lower troposphere (925, 850, and 700 hPa) the percentage error in specific humidity for a temperature uncertainty of 1 K was less than 3 % in the tropics, 6 % in middle latitudes, and 10 % in high latitudes.At 400 hPa, the percent errors grew to 18 % in the tropics, 45 % in middle latitudes, and 67 % in high latitudes.

Figure 1 .
Figure 1.Profiles extracted from the CONTRAST flights during the experiment.

Figure 4 .
Figure 4. Two snapshots of the troposphere for (top) 19 February 2014 around 18:00 UT and (bottom) 20 February 2014 around 00:00 UT.Left panels: profiles for CONTRAST, RO, ERA, GFS, RS, and AIRS for the 2 days.Right panels: ERA RH (%, color shading) at 500 hPa, ozone values (white contours: values ≥ 50 ppbv are solid, otherwise dashed), and winds (black arrows).The white X marks the location of Guam.The locations of CONTRAST, RO, RS, and AIRS profiles are marked.Additional squares indicate other RO measurements.The squares marking the locations of the RO and CONTRAST locations are colored according to the RH of each profile at the 500 hPa level (matching the color bar).

Figure 6 .
Figure 6.Scatter plots for N (top left), T (top right), q (bottom left, logarithmic scale), and RH (bottom right) for all 3 h 600 km collocation pairs for RO and CONTRAST.The color indicates the altitude of the measurement.

Figure 7 .
Figure 7. Scatter plots for RH for all 3 h 600 km collocation pairs for (left) ERA and CONTRAST on the 60 s smoothed CONTRAST altitude grid and (right) GFS and ERA on the ERA (pressure) grid.The color indicates the altitude of the measurement (left: km; right: hPa).

Figure 8 .
Figure 8. Frequency for RH < 10 % at the potential temperature level 320 K for DJF and JJA 2014.
4. When compared to CONTRAST, RO has a moist bias for low humidity values and a dry bias for high humidity values.Similar results are found when comparing RO to long records of tropical RS measurements (Appendix A).

Figure A3 .
Figure A3.Specific humidity from the RO 1D-Var retrieval compared to the RO simple retrieval, using the same profiles as in Fig.A2.Points from profiles for which the collocated radiosonde profile experiences super-refraction are marked by an X.

Figure B1 .
Figure B1.RH, T , q, and e profiles with the original ERA first guess (dashed blue), the original 1D-Var output (solid blue), the "high moisture" first guess (dashed, shades of red), and the 1D-Var output for the modified first guess (solid, shades of red).

Table 2 .
Mean, RMS, and Pearson R coefficients for differences in relative humidity of RO-CONTRAST, ERA-CONTRAST, and GFS-CONTRAST for three different collocation criteria.